fix: improve perf of instructions w/ rotr

This commit is contained in:
Rekai Nyangadzayi Musuka 2022-03-16 22:28:24 -03:00
parent 1921218c7b
commit 39ab363afa
7 changed files with 33 additions and 13 deletions

View File

@ -5,6 +5,7 @@ const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
const InstrFn = @import("../../cpu.zig").ArmInstrFn; const InstrFn = @import("../../cpu.zig").ArmInstrFn;
const sext = @import("../../util.zig").sext; const sext = @import("../../util.zig").sext;
const rotr = @import("../../util.zig").rotr;
pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I: bool, comptime W: bool, comptime L: bool) InstrFn { pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I: bool, comptime W: bool, comptime L: bool) InstrFn {
return struct { return struct {
@ -38,7 +39,7 @@ pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I:
0b01 => { 0b01 => {
// LDRH // LDRH
const value = bus.read16(address & 0xFFFF_FFFE); const value = bus.read16(address & 0xFFFF_FFFE);
result = std.math.rotr(u32, value, 8 * (address & 1)); result = rotr(u32, value, 8 * (address & 1));
}, },
0b10 => { 0b10 => {
// LDRSB // LDRSB
@ -52,7 +53,7 @@ pub fn halfAndSignedDataTransfer(comptime P: bool, comptime U: bool, comptime I:
break :blk sext(16, bus.read16(address)); break :blk sext(16, bus.read16(address));
}; };
result = std.math.rotr(u32, value, 8 * (address & 1)); result = rotr(u32, value, 8 * (address & 1));
}, },
0b00 => unreachable, // SWP 0b00 => unreachable, // SWP
} }

View File

@ -7,6 +7,8 @@ const PSR = @import("../../cpu.zig").PSR;
const log = std.log.scoped(.PsrTransfer); const log = std.log.scoped(.PsrTransfer);
const rotr = @import("../../util.zig").rotr;
pub fn psrTransfer(comptime I: bool, comptime R: bool, comptime kind: u2) InstrFn { pub fn psrTransfer(comptime I: bool, comptime R: bool, comptime kind: u2) InstrFn {
return struct { return struct {
fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void { fn inner(cpu: *Arm7tdmi, _: *Bus, opcode: u32) void {
@ -22,7 +24,7 @@ pub fn psrTransfer(comptime I: bool, comptime R: bool, comptime kind: u2) InstrF
// MSR // MSR
const field_mask = @truncate(u4, opcode >> 16 & 0xF); const field_mask = @truncate(u4, opcode >> 16 & 0xF);
const rm_idx = opcode & 0xF; const rm_idx = opcode & 0xF;
const right = if (I) std.math.rotr(u32, opcode & 0xFF, (opcode >> 8 & 0xF) << 1) else cpu.r[rm_idx]; const right = if (I) rotr(u32, opcode & 0xFF, (opcode >> 8 & 0xF) << 1) else cpu.r[rm_idx];
if (R and !cpu.hasSPSR()) log.warn("Tried to write to SPSR in User/System Mode", .{}); if (R and !cpu.hasSPSR()) log.warn("Tried to write to SPSR in User/System Mode", .{});

View File

@ -4,6 +4,8 @@ const Bus = @import("../../Bus.zig");
const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi; const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
const InstrFn = @import("../../cpu.zig").ArmInstrFn; const InstrFn = @import("../../cpu.zig").ArmInstrFn;
const rotr = @import("../../util.zig").rotr;
pub fn singleDataSwap(comptime B: bool) InstrFn { pub fn singleDataSwap(comptime B: bool) InstrFn {
return struct { return struct {
fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void { fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
@ -20,7 +22,7 @@ pub fn singleDataSwap(comptime B: bool) InstrFn {
cpu.r[rd] = value; cpu.r[rd] = value;
} else { } else {
// SWP // SWP
const value = std.math.rotr(u32, bus.read32(address & 0xFFFF_FFFC), 8 * (address & 0x3)); const value = rotr(u32, bus.read32(address & 0xFFFF_FFFC), 8 * (address & 0x3));
bus.write32(address & 0xFFFF_FFFC, cpu.r[rm]); bus.write32(address & 0xFFFF_FFFC, cpu.r[rm]);
cpu.r[rd] = value; cpu.r[rd] = value;
} }

View File

@ -6,6 +6,8 @@ const Bus = @import("../../Bus.zig");
const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi; const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
const InstrFn = @import("../../cpu.zig").ArmInstrFn; const InstrFn = @import("../../cpu.zig").ArmInstrFn;
const rotr = @import("../../util.zig").rotr;
pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool, comptime B: bool, comptime W: bool, comptime L: bool) InstrFn { pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool, comptime B: bool, comptime W: bool, comptime L: bool) InstrFn {
return struct { return struct {
fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void { fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u32) void {
@ -33,7 +35,7 @@ pub fn singleDataTransfer(comptime I: bool, comptime P: bool, comptime U: bool,
} else { } else {
// LDR // LDR
const value = bus.read32(address & 0xFFFF_FFFC); const value = bus.read32(address & 0xFFFF_FFFC);
result = std.math.rotr(u32, value, 8 * (address & 0x3)); result = rotr(u32, value, 8 * (address & 0x3));
} }
} else { } else {
if (B) { if (B) {

View File

@ -3,6 +3,8 @@ const std = @import("std");
const Arm7tdmi = @import("../cpu.zig").Arm7tdmi; const Arm7tdmi = @import("../cpu.zig").Arm7tdmi;
const CPSR = @import("../cpu.zig").PSR; const CPSR = @import("../cpu.zig").PSR;
const rotr = @import("../util.zig").rotr;
pub fn execute(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 { pub fn execute(comptime S: bool, cpu: *Arm7tdmi, opcode: u32) u32 {
var result: u32 = undefined; var result: u32 = undefined;
if (opcode >> 4 & 1 == 1) { if (opcode >> 4 & 1 == 1) {
@ -141,7 +143,7 @@ pub fn arithmeticRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8)
} }
pub fn rotateRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 { pub fn rotateRight(comptime S: bool, cpsr: *CPSR, rm: u32, total_amount: u8) u32 {
const result = std.math.rotr(u32, rm, total_amount); const result = rotr(u32, rm, total_amount);
if (S and total_amount != 0) { if (S and total_amount != 0) {
cpsr.c.write(result >> 31 & 1 == 1); cpsr.c.write(result >> 31 & 1 == 1);

View File

@ -4,6 +4,8 @@ const Bus = @import("../../Bus.zig");
const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi; const Arm7tdmi = @import("../../cpu.zig").Arm7tdmi;
const InstrFn = @import("../../cpu.zig").ThumbInstrFn; const InstrFn = @import("../../cpu.zig").ThumbInstrFn;
const rotr = @import("../../util.zig").rotr;
pub fn format6(comptime rd: u3) InstrFn { pub fn format6(comptime rd: u3) InstrFn {
return struct { return struct {
fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void { fn inner(cpu: *Arm7tdmi, bus: *Bus, opcode: u16) void {
@ -39,7 +41,7 @@ pub fn format78(comptime op: u2, comptime T: bool) InstrFn {
0b10 => { 0b10 => {
// LDRH // LDRH
const value = bus.read16(address & 0xFFFF_FFFE); const value = bus.read16(address & 0xFFFF_FFFE);
cpu.r[rd] = std.math.rotr(u32, value, 8 * (address & 1)); cpu.r[rd] = rotr(u32, value, 8 * (address & 1));
}, },
0b11 => { 0b11 => {
// LDRSH // LDRSH
@ -49,7 +51,7 @@ pub fn format78(comptime op: u2, comptime T: bool) InstrFn {
break :blk sext(16, bus.read16(address)); break :blk sext(16, bus.read16(address));
}; };
cpu.r[rd] = std.math.rotr(u32, value, 8 * (address & 1)); cpu.r[rd] = rotr(u32, value, 8 * (address & 1));
}, },
} }
} else { } else {
@ -66,7 +68,7 @@ pub fn format78(comptime op: u2, comptime T: bool) InstrFn {
0b10 => { 0b10 => {
// LDR // LDR
const value = bus.read32(address & 0xFFFF_FFFC); const value = bus.read32(address & 0xFFFF_FFFC);
cpu.r[rd] = std.math.rotr(u32, value, 8 * (address & 0x3)); cpu.r[rd] = rotr(u32, value, 8 * (address & 0x3));
}, },
0b11 => { 0b11 => {
// LDRB // LDRB
@ -93,7 +95,7 @@ pub fn format9(comptime B: bool, comptime L: bool, comptime offset: u5) InstrFn
// LDR // LDR
const address = cpu.r[rb] + (@as(u32, offset) << 2); const address = cpu.r[rb] + (@as(u32, offset) << 2);
const value = bus.read32(address & 0xFFFF_FFFC); const value = bus.read32(address & 0xFFFF_FFFC);
cpu.r[rd] = std.math.rotr(u32, value, 8 * (address & 0x3)); cpu.r[rd] = rotr(u32, value, 8 * (address & 0x3));
} }
} else { } else {
if (B) { if (B) {
@ -121,7 +123,7 @@ pub fn format10(comptime L: bool, comptime offset: u5) InstrFn {
if (L) { if (L) {
// LDRH // LDRH
const value = bus.read16(address & 0xFFFF_FFFE); const value = bus.read16(address & 0xFFFF_FFFE);
cpu.r[rd] = std.math.rotr(u32, value, 8 * (address & 1)); cpu.r[rd] = rotr(u32, value, 8 * (address & 1));
} else { } else {
// STRH // STRH
bus.write16(address & 0xFFFF_FFFE, @truncate(u16, cpu.r[rd])); bus.write16(address & 0xFFFF_FFFE, @truncate(u16, cpu.r[rd]));
@ -139,7 +141,7 @@ pub fn format11(comptime L: bool, comptime rd: u3) InstrFn {
if (L) { if (L) {
// LDR // LDR
const value = bus.read32(address & 0xFFFF_FFFC); const value = bus.read32(address & 0xFFFF_FFFC);
cpu.r[rd] = std.math.rotr(u32, value, 8 * (address & 0x3)); cpu.r[rd] = rotr(u32, value, 8 * (address & 0x3));
} else { } else {
// STR // STR
bus.write32(address & 0xFFFF_FFFC, cpu.r[rd]); bus.write32(address & 0xFFFF_FFFC, cpu.r[rd]);

View File

@ -1,12 +1,21 @@
const std = @import("std"); const std = @import("std");
const Log2Int = std.math.Log2Int;
pub fn sext(comptime bits: comptime_int, value: u32) u32 { pub inline fn sext(comptime bits: comptime_int, value: u32) u32 {
comptime std.debug.assert(bits <= 32); comptime std.debug.assert(bits <= 32);
const amount = 32 - bits; const amount = 32 - bits;
return @bitCast(u32, @bitCast(i32, value << amount) >> amount); return @bitCast(u32, @bitCast(i32, value << amount) >> amount);
} }
/// See https://godbolt.org/z/W3en9Eche
pub inline fn rotr(comptime T: type, value: T, r: anytype) T {
comptime std.debug.assert(@typeInfo(T).Int.signedness == .unsigned);
const ar = @truncate(Log2Int(T), r);
return value >> ar | value << @truncate(Log2Int(T), @typeInfo(T).Int.bits - @as(T, ar));
}
pub const FpsAverage = struct { pub const FpsAverage = struct {
const Self = @This(); const Self = @This();